Sustainability 13 08578 v5

sustainability
Article
COVID-19 Pandemic: Is the Crypto Market a Safe Haven?
The Impact of the First Wave
Darko Vukovic 1,2,3 , Moinak Maiti 4, * , Zoran Grubisic 5 , Elena M. Grigorieva 1,2 and Michael Frömmel 6
1 International Laboratory for Finance and Financial Markets Finance, Faculty of Economics,
People’s Friendship University of Russia (RUDN University), Miklukho-Maklaya Str.6, 117198 Moscow,
Russia; vdarko@hotmail.rs (D.V.); aroooveo@yandex.ru (E.M.G.)
2 Finance and Credit Department, Faculty of Economics, People’s Friendship University of Russia
(RUDN University), Miklukho-Maklaya Str.6, 117198 Moscow, Russia
3 Geographical Institute “Jovan Cvijic” SASA, Djure Jaksica 9, 11000 Belgrade, Serbia
4 National Research University Higher School of Economics, St. Petersburg School of Economics and
Management, Department for Finance, Kantemirovskaya St. 3A, 194100 Sankt Petersburg, Russia
5 Belgrade Banking Academy, Faculty for Banking, Insurance and Finance, Zmaj Jovina 12,
11000 Belgrade, Serbia; zoran.grubisic@bba.edu.rs
6 Department of Economics, Ghent University, Sint-Pietersplein 5, 9000 Gent, Belgium;
michael.froemmel@UGent.be
* Correspondence: mmaiti@hse.ru
Abstract: The present study investigated whether the crypto market is a safe haven. The study
argues that during the first wave of the COVID-19 crisis, gold and oil, as typical global commodities,
could have been diversifiers. The study developed a unique COVID-19 global composite index that

measures COVID-19 pandemic time-variant movements on each day. The study used OLS (ordinary
least squares), quantile, and robust regressions to check whether the COVID-19 crisis has had any
Citation: Vukovic, D.; Maiti, M.;
significant direct influence on the crypto market. The OLS, quantile, and robust regressions estimates
Grubisic, Z.; Grigorieva, E.M.;
Frömmel, M. COVID-19 Pandemic:
confirmed that there was no statistically significant direct influence of the COVID-19 crisis on the
Is the Crypto Market a Safe Haven? crypto market in the first wave period. However, the study found spillovers from risky assets (S&P
The Impact of the First Wave. 500) on the crypto market, with Tether as an exception. Due to this special characteristic, Tether
Sustainability 2021, 13, 8578. https:// might present a safe haven within the crypto market.
doi.org/10.3390/su13158578
Keywords: cryptocurrency; COVID-19; safe haven; quantile regression; tether
Academic Editor: Antonio Boggia
Received: 25 June 2021

Accepted: 26 July 2021 1. Introduction
Published: 31 July 2021
The coronavirus disease of 2019 (COVID-19) was detected in Wuhan, China at the
Corrected: 12 November 2021
beginning of December 2019. However, it was officially reported for the first time to the
world by the World Health Organization (WHO 2020) on 31 December 2019. Within a
Publisher’s Note: MDPI stays neutral
month, WHO declared the COVID-19 outbreak as a public health emergency of interna-
with regard to jurisdictional claims in
published maps and institutional affil-
tional concern at their second meeting of the Emergency Committee (WHO, 2020). The
iations.
virus had already spread to all parts of the world, and on 11 March 2020, WHO declared
it as a global pandemic. Fear and panic did not bypass the world economy. The Dow
Jones Industrial Average Index downscaled more than 40% from 29.348 on 19 February
to 18.591 on 23 March 2020 (MarketWatch 2020). A similar trend was observed for S&P
500 (downscaled from 3.386 to 2.237) and other well-known world indices during the same
Copyright: © 2021 by the authors.
period.
Licensee MDPI, Basel, Switzerland.
The global scientific and business community believes that the world economy will
This article is an open access article
distributed under the terms and
have one of the biggest crises ever. Financial markets, airline industries, SMEs, tourism
conditions of the Creative Commons
and hospitality, services, construction, and mortgage sectors all recorded a large drop in
Attribution (CC BY) license (https://
the first quarter of 2020. According to [1], this crisis is a “Black Swan” event, as originally
creativecommons.org/licenses/by/ described by [2]. The study by [1] argued that this event was expected to have an extreme
4.0/).
Sustainability 2021, 13, 8578. https://doi.org/10.3390/su13158578 https://www.mdpi.com/journal/sustainability

Sustainability 2021, 13, 8578 2 of 17
impact that lies outside the area of regular expectations. Several studies have focused on
the existence of this issue during crises [3–5] and epidemic diseases [6,7].
The global demand for many goods and services has plummeted due to a large portion
of the population is in quarantine across the globe. The government imposed strict social
distancing measures in an attempt to prevent the spread of the disease. The recession
has hit many sectors and businesses. In the first quarter of 2020, the world oil and gas
sectors faced a deep recession due to its falling demand caused by the ongoing pandemic
and policy failures of the OPEC countries. On 20 April 2020, the West Texas Intermediate
futures (WTI futures) contract had slumped from $17.85 a barrel to −$37.63 (Bloomberg,
2020). For the first time in history, oil futures became negative and oil prices dropped
below 1985’s minimum. However, gold shows a different trend. According to Bloomberg,
after a short fall in gold’s price in early March ($1500 and less per ounce). The gold price
rose continuously and exceeded $1700 within a month after. The Bank of America raised
its 18-month gold price target to $3000 (Bloomberg).
It is important to look at how cryptocurrencies behave in this situation. The greatest
challenge that lies in the context of the ongoing COVID-19 crisis is to find a “safe haven”.
A safe haven is supposed to be immune to this crisis and protect the value of assets of
investors. Several studies have shown that cryptocurrencies (especially Bitcoin) have
no hedging or safe haven capabilities [8,9]. According to the study of [10], Bitcoin has
fewer characteristics as a hedge and safe haven compared to gold. Moreover, it is not
the best instrument to hedge structural crude oil price shocks (compared to gold) [11].
Cryptocurrencies showed a higher correlation with riskier assets between the period of
March and April 2020 (Bloomberg). Bitcoin slumped from $8000 to −$4000 in a few days
in mid-March 2020. However, within a month, Bitcoin hit strong resistance around $7500,
as market players reacted positively to the news that COVID-19 outbreaks had started to
stabilize in some of the worst affected countries. This is in accordance with the authors
of [12], who stated that Bitcoin is a highly volatile asset with extreme price dynamics
and price clustering. It shows a positive correlation with downward markets [10]. This
unique situation will stimulate numerous studies in the future to test the speculative
activity of investors and their real influence on global market prices [13]. According to
preliminary data, investors have no strong speculative impact on global market prices
during a pandemic crisis.
The present study was inspired by the following two questions: “Is there any safe
haven for cryptocurrencies during a pandemic crisis?”, and “Which cryptocurrency can be
used for hedging?” The study investigated the connection of cryptocurrencies to gold and
oil during the first wave of the pandemic (the first 6 months) along with the response of
financial markets to these variables. To examine the impact, the study used mean-based
OLS regression, conditional median-based quantile regression, and robust regression for
additional robustness checks. Moreover, depending on the regression results, the study
applied the model that was most suitable for the tested variables (in the case of high
tail value estimations). The study also examined the co-movement and contagion effect
between cryptocurrency pairs using wavelet coherence diagrams. Thus, the goal of the
study was to examine all such interdependencies that arose (in the first wave of the crisis)
due to the ongoing COVID-19 crisis to support future studies. The COVID-19 crisis seems
to be ongoing and is about to reach its three highest peaks across the globe. The present
study developed a composite COVID-19 index to answer the fundamental question “How
should one measure the effect of COVID-19 globally?” The study findings have both
scientific and practical significance, bearing in mind the development of financial theory in
the field of crises and cryptocurrencies (theoretical), on the one hand, and investor reactions
to the shocks (practical) on the other. The study adds value to the existing literature by
creating a COVID-19 global composite index and examining the impact of the ongoing
COVID-19 crisis on the crypto market. More specifically, the present study contributes by
investigating more specific crypto market issues, such as if the crypto market is a hedge or
safe haven.
2. Theoretical Background
The influence of the COVID-19 crisis on cryptocurrency has been studied by several
authors in the last few months [6,12]. The study by [6] considered five cryptocurrencies to
investigate herding biases by quantifying the self-similarity intensity of cryptocurrency
returns’ during the COVID-19 pandemic. The study found that the COVID-19 crisis has
had a positive impact on the efficiency of the cryptocurrency market. To test the influence of
the COVID-19 pandemic on cryptocurrency market efficiency, the study used the following
methodologies: a multifractal detrended fluctuation approach, the Magnitude of Long
memory index, and the generalized Hurst exponent. The study also highlighted that
Bitcoin was more efficient before the COVID-19 crisis. Likewise, Ethereum was more
efficient during and after the outbreak of the COVID-19 crisis. The study by [14] analyzed
the impact of the COVID-19 crisis on 15 equity indices, 4 bond benchmark indices, 9
precious metals, and 3 cryptocurrencies (Bitcoin, Ethereum, and Litecoin) using a quantile
autoregression model. The study found that the COVID-19 crisis could be a “Black Swan”
event, especially for cryptocurrencies. Corbet et al. [12] found that there is negative
sentiment toward COVID-19 and cryptocurrency returns. The study also highlighted
that investors could have benefitted from the investment diversification, as digital assets
(social media) acted as a safe haven in comparison to precious metals during previous
crises (similar to the case of a “Black Swan”). The study used a standard GARCH model
(generalized autoregressive conditional heteroskedasticity) and the Python package to
derive the above inference. The study by [15,16] forecasts the impact of COVID-19 will be
significant, both economically and socially. However, the study does not entirely support
the findings of the previous studies that the COVID-19 pandemic can be seen as a “Black
Swan” event. This is due to the fact that this crisis is totally unexpected, with strong
impacts not only on the global economy and finance but also a health crisis. In the study
of [17], wavelet methods were applied to examine the impact of COVID-19 on Bitcoin
prices. The study used daily data of COVID-19 world deaths and daily Bitcoin prices
from 31 December 2019 to 29 April 2020. The study found that the COVID-19 pandemic
affected the price growth of Bitcoin. Hereafter, this section additionally discusses the
theoretical explanations concerning cryptocurrency behavior, methodologies, and studies
that analyzed precious metals and cryptocurrencies.
The crypto market is very volatile, especially with its most famous currency, Bitcoin.
Bitcoin is mostly used for speculative purposes, which causes high volatility and market
bubbles [17–19]. Given the growing interest in Bitcoin, it is important to choose a reliable
model to forecast the risk of such an investment. By using specifications that can account
for structural breaks in GARCH, namely Markov switching GARCH models, the authors
of [20,21] analyzed Bitcoin daily log returns exhibiting regime changes in their volatility
dynamics. Because of such dynamics, it is important to explore ways to reduce the level of
risk of an investment in Bitcoin and to hedge or discover some factors that can influence
the price movement of Bitcoin.
Tether is the only important stable coin on the crypto market with significant market
capitalization. Tether is purportedly backed by USD (US dollar) reserves, but there is
no clear evidence of it. The study by [22] tested two hypotheses regarding the role of a
Tether in the crypto world, especially on Bitcoin. The first hypothesis put forward was that
demand is driven by the idea that Tether is being used as a medium of exchange for fiat
currency’s entry into the crypto world. The second hypothesis was that Tether flows cause
positive Bitcoin returns. The obvious argument behind this is that Tether is being printed,
without support, in USD, and is pushed out into the market with an inflationary effect on
asset prices. This is in accordance with the study of [22], which has shown that none of
the exposures to macroeconomic factors, stocks markets, currencies, or commodities can
explain cryptocurrency prices. On the other hand, ref. [20] found in their study that Tether
has a sizable impact on Bitcoin prices. These findings are generally consistent with the
evidence that sophisticated investors may earn profit from the bubbles [23].
This is interesting due to the fact that [24–26] claim that in some situations, the spec-
ulative activity of investors will not be the main reason for volatilities on global market
prices. Similarly, according to [18,27,28], cryptocurrencies are characterized by extreme
volatilities and bubbles, price jumps [29], co-explosivity [30], and strong speculative targets
(especially Bitcoin, Ethereum, and similar cryptocurrencies). However, cryptocurrencies
are not only currencies but also assets [28,31,32] and an investment destination for in-
vestors [12]. The study by [33] analyzed returns and volatility spillovers between Bitcoin
and four major global asset classes (stocks, commodities, currencies, and bonds). The
study used the VAR GARCH-in-mean model (value-at-risk in generalized autoregressive
conditional heteroskedasticity) and concluded that the Bitcoin market is not completely
isolated. The sign of spillovers exhibited some differences in the two market conditions,
with greater evidence that Bitcoin receives more volatility than it transmits.
As Bitcoin and gold are often looked at as safe havens, the relationship between these
assets is important for discovering if there is any hedge option between them. There is
always a question of the hedging capabilities of Bitcoin regarding some stock exchange
indices. Baek and Elbeck [32] explored the financial asset capabilities of Bitcoin using
GARCH models. The initial model showed several similarities to gold and the USD,
indicating hedging capabilities and advantages as a medium of exchange. By applying the
asymmetric GARCH methodology [34], Bitcoin can be used to hedge its position against
FTSE 100, and in the short term, against the USD. In another study, ref. [35] based their
findings on this original sample and an extended sample period. The study showed that
Bitcoin exhibits distinctively different return, volatility, and correlation characteristics
compared to other assets, including gold and the USD. The study by [9] proposed new
definitions of weak and strong safe havens within a bivariate cross-quantilogram approach
to explore whether Bitcoin can improve hedging strategies for stock market investments
during extreme market conditions, and to examine whether this function is similar to gold
hedging strategies. The study results show that Bitcoin and gold play the role of a weak
safe haven asset in a few cases. Besides gold, oil has always played a significant role in the
global commodity market in investors’ hedging strategies against stock exchange indices.
The study by [36] used the GARCH model based on 5 years of sample data to explore
Bitcoin’s volatility effect on crude oil and gold. The study found that Bitcoin volatility had
a negative impact on gold and crude oil volatility. The study’s estimates clearly indicate
that Bitcoin and gold can be used as hedging instruments.
Gold has always played an important role in the global commodity market. The
explanation behind the global economy today is that gold is an inflation hedge. It is
commonly believed that in a time of crisis, gold is a typical example of a safe haven.
However, it can always be put into question from a dynamic point of view; is there only
one safe haven for all time? The study by [37,38] analyzed the dynamic role of gold as a
safe haven and found that investors’ behavior can destroy this role in practice.
The analysis by [39] seems to support this finding by applying a VAR-ADCC-BVGARCH
(vector autoregressive asymmetric dynamic conditional correlation bivariate generalized
autoregressive conditional heteroskedasticity) to the US financial market during the pe-
riod of 2007–2017, although there is consensus in the literature that gold could be a risk-
diminishing instrument against stocks. The findings of the previous studies suggest that
the hedging characteristic of gold tends to mitigate against US stocks as market capital-
ization increases, implying there is a large proportion of funds invested in gold, against
stocks. The study by [39] examined the hedge or safe haven role of the gold relative to
the Dow Jones stock industry indices by using quantile GARCH analysis. However, the
study findings are dubious depending on the time frame, whether it is the whole sample
period (1980–2017) or each sub-period separately (1980–1995 and 1996–2017). Furthermore,
it differs across different sectors. The study by [40] applied different methodologies in
their robust analysis, indicating the weak hedge and safe haven role of gold for stocks.
The study by [41] applied a wavelet decomposition with a copula approach to analyze the
dependence between returns of gold and other assets (bonds, stocks, and exchange rates).
The study found that everything changed after the 2008 crisis. Before the crisis, gold was
able to shield investors. However, in the aftermath, gold is unable to serve as a hedge or
safe haven in the classical sense.
According to the existing literature, most of the studies measure volatility and con-
nectedness using the GARCH models. Some of the very recent studies found that GARCH
models are not appropriate for tail values of skewed distribution. Similarly, many stud-
ies [42–45] argue that tail risk modeling is very important to consider in cryptocurrency
value regressions. For example, the authors of [44] suggest a bivariate copula approach
for tail risk modeling, the authors of [44] propose a linear bivariate quantile regression
(bivariate CoVaR) as the most appropriate model, and [43] (based on the work of [46,47]),
used a tail-event driven network to analyze the tail-risk interdependence among 23 cryp-
tocurrencies. One of the biggest advantages of such models in tail value measures is in the
description of the complexity of the crypto market, especially for the conditional tail-risk
and tail-event driven networks used by [43,44]. Most importantly, it turns the distributional
effect of the independent variables on the dependent variable into different quantile ranges.
According to [48], the most significant characteristic of quantile regression is to make a
classification of the heterogeneous effects of heterogeneous cross-sectional ranges. The
median is a more appropriate measure than the mean in the case of conditional quantile
estimations [49,50]. This technique is the most effective in tail value estimations (extremes
of a dataset).
In estimations of the mean value of the dependent variable for particular levels of the
independent variables, generally, the OLS (ordinary least square) [51,52] is used. The OLS
is an appropriate model for the conditional mean function to describe how the mean of the
response variable changes with the covariates [53]. The main limitations of the OLS are (1)
that it is often confronted with the heteroscedasticity, focusing on the means as measured
facts for tails values are lost, (2) it fails to recognize the source and nature of heterogeneity,
and (3) it is inappropriate for extreme observations. Studies by [49,50] argue that quantile
regression provides essential information about the relationship between the response
variable and covariates on the entire conditional distribution. According to [53–56], such
a model visualizes heteroscedasticity in the dataset. Similarly, the model determines the
influence of covariates on the shape and scale of the entire response distribution [53].
Quantile regression models are mostly used in portfolio studies and different financial
analyses. This technique was popularized by [53], who found that it is possible to model the
relationship between returns and beta for companies that overperform and underperform
relative to the mean. In the study of [57], it was shown that quantile regression is the most
appropriate model in the study of the sensitivity of cryptocurrency returns to changes
in gold price returns. A similar technique was previously used by [44] to study hedging
strategies against global uncertainty based on cryptocurrency. Great attention has been
paid to the tail behavior of values in their studies.
3. Data and Methodology

3.1. Data
The study dataset consisted of the following variables for studying the impact of
the first wave of COVID-19 between 22 January 2020 and 11 April 2020: daily returns of
five cryptocurrencies (Bitcoin, Ethereum, XRP, Tether, and Bitcoin Cash) (data source:
https://coinmarketcap.com/) (accessed on 15 April 2020), google trends of searches (data
source: Google); gold daily returns (data source: Yahoo); crude oil daily returns (Data
Source: Yahoo); VIX index daily returns (Data Source: Yahoo); S&P 500 index daily returns
(Data Source: Yahoo), and COVID-19 global index (data source: https://data.humdata.org/
dataset/novel-coronavirus-2019-ncov-cases) (accessed on 15 April 2020). All daily returns
were calculated using the HPR (Holding Period Returns) formula. To capture the time-
varying effect of COVID-19, the study created the COVID-19 global index (explained in
detail in Section 3.2). The study variables were as follows: (a) the dependent variable is the
cryptocurrency daily average returns; (b) the independent variable was COVID-19 global
index (COVID-19), developed to measure the effect of COVID-19 and daily movements
of COVID-19; GOLD: important global commodity; GT: Google trend on how much a
particular cryptocurrency was searched during the study period; OIL: another important
global commodity, the S&P 500 index is a good indicator of how financial markets are
doing; VIX is intended to provide an instantaneous volatility measure regarding how much
the market believes that the market index will fluctuate in the short term.
3.2. Methodology
3.2.1. COVID-19 Global Index Construction
For the present study, we constructed a composite COVID-19 index following [58–60]
to answer the fundamental question of how one should measure the effect of COVID-19
globally. The construction of the composite COVID-19 global index requires three stages.
The first stage of the composite COVID-19 global index development started with the
normalization of the data points, followed by a weighting of the data points and weighted
aggregates of three variables: confirmed cases of COVID-19; deaths due to COVID-19, and
recovered cases of COVID-19. We used z-scores and rescaling methods for normalization
with the following expression:
−
x−x
z=
s
Then, the rescaling method was deployed on all the normalized data for a linear
transformation of the data into a scale of 1–100. The 1–100 scale was determined by the
following equation:
x − min( x )
y=
max( x ) − min( x )
We used these steps due to the fact that the effects of COVID-19 cannot be measured
by a single indicator and requires the composite index used by the World Bank and the UN.
Likewise, we normalized and standardized the data by creating a COVID-19 global index.
The COVID-19 global index was constructed from three weighted components—confirmed
cases of COVID-19; deaths due to COVID-19 and recovered cases of COVID-19—which
were collected from the humanitarian data exchange (https://data.humdata.org/dataset/
novel-coronavirus-2019-ncov-cases) (accessed on 15 April 2020). The dataset consisted of
81 daily observations from 22 January 2020 to 11 April 2020.
There is a high probability that a high level of multicollinearity exists among these
three variables that are important measures of the impact of COVID-19. It is also true that
the weights of these three variables are not uniform across all countries. Such issues could
result in a distortion of the overall calculation of these variables. To avoid such errors in this
estimation study, we normalized and standardized the data by constructing a COVID-19
global index. All these three variables were converted to a scale of 1–100 points, where 100
denotes the best score and 1 denotes the worst. Equal weights were assigned to all three
variables in preparation for the aggregate COVID-19 global index. Figure 1a shows the
daily movements of the independent variables. The primary axis represents the COVID
and VIX daily movements, and the secondary axis represents the GOLD, OIL, and S&P
500 daily movements. Figure 1b shows the daily movements of the dependent variables.
The Tether daily movements are represented on the secondary axis and the rest are on the
primary axis.
Sustainability 2021, 13, x FOR PEER REVIEW 7 of 18
(a)
2.5 0.3
2 0.2
0.1
1.5
0
1
-0.1
0.5
-0.2
0 -0.3
Jan 22, 2020
Jan 24, 2020
Jan 26, 2020
Jan 28, 2020
Jan 30, 2020
Feb 01, 2020
Feb 03, 2020
Feb 05, 2020
Feb 07, 2020
Feb 09, 2020
Feb 11, 2020
Feb 13, 2020
Feb 15, 2020
Feb 17, 2020
Feb 19, 2020
Feb 21, 2020
Feb 23, 2020
Feb 25, 2020
Feb 27, 2020
Feb 29, 2020
Mar 02, 2020
Mar 04, 2020
Mar 06, 2020
Mar 08, 2020
Mar 10, 2020
Mar 12, 2020
Mar 14, 2020
Mar 16, 2020
Mar 18, 2020
Mar 20, 2020
Mar 22, 2020
Mar 24, 2020
Mar 26, 2020
Mar 28, 2020
Mar 30, 2020
Apr 01, 2020
Apr 03, 2020
Apr 05, 2020
Apr 07, 2020
Apr 09, 2020
Apr 11, 2020
-0.5 -0.4
COVID VIX GOLD OIL S&P 500

(b)
0.3 0.06
0.2
0.04
0.1
0.02
0
Feb 01, 2020
Feb 03, 2020
Feb 05, 2020
Feb 07, 2020
Feb 09, 2020
Feb 11, 2020
Feb 13, 2020
Feb 15, 2020
Feb 17, 2020
Feb 19, 2020
Feb 21, 2020
Feb 23, 2020
Feb 25, 2020
Feb 27, 2020
Feb 29, 2020
Mar 02, 2020
Mar 04, 2020
Mar 06, 2020
Mar 08, 2020
Mar 10, 2020
Mar 12, 2020
Mar 14, 2020
Mar 16, 2020
Mar 18, 2020
Mar 20, 2020
Mar 22, 2020
Mar 24, 2020
Mar 26, 2020
Mar 28, 2020
Mar 30, 2020
Jan 22, 2020
Jan 24, 2020
Jan 26, 2020
Jan 28, 2020
Jan 30, 2020
Apr 01, 2020

Apr 03, 2020
Apr 05, 2020
Apr 07, 2020
Apr 09, 2020
Apr 11, 2020
-0.1 0
-0.2
-0.02
-0.3
-0.04
-0.4
-0.5 -0.06
Bitcoin Etherum XRP BitcoinCash Tether
Figure
Figure (a,b):
1. (a)1.and Variables’
(b): daily
Variables’ movements.
daily movements.
3.2.2. Wavelet Coherence
3.2.2. Wavelet Coherence
To study the co-movement between the cryptocurrencies’ daily returns, a wavelet
To study the co-movement between the cryptocurrencies’ daily returns, a wavelet
coherence diagram was plotted using the “biwavelet” R package, similar to the studies
coherence diagram was plotted using the “biwavelet” R package, similar to the studies of
of [61,62]. We analyzed the continuous co-movement between the pairs at different times
[60,61]. We analyzed the continuous co-movement between the pairs at different times
and frequency domains using a wavelet coherence diagram. The wavelet coherence could
and frequency domains using a wavelet coherence diagram. The wavelet coherence could
be perceived as a measure of a local correlation of the two time series, both in time and
be perceived as a measure of a local correlation of the two time series, both in time and
scale. This procedure is analogous to coherence in Fourier analysis. It is defined as the
scale. This procedure is analogous to coherence in Fourier analysis. It is defined as the
squared absolute value of the smoothed cross wavelet spectra normalized by the product
squared absolute value of the smoothed cross wavelet spectra normalized by the product
of the smoothed individual wavelet power spectra of each series:
of the smoothed individual wavelet power spectra of each series:
S s Wxy (u, s) 2
−1
2
) =s) =
R (u,R s(u, S(s W (u, s)) (1)

−1 |W (u, s)|2 S s−1 W (u, s) 2
(1)
S sS(s |W (u, s)| )S(s W (u, s) )
x y
3.2.3. Regressions
3.2.3. Regressions
We deployed two versions of regressions: (1) mean-based OLS regression, and (2)
We deployed
conditional two versions
median-based of regression.
quantile regressions: (1) mean-based OLS regression, and (2)
conditional median-based quantile regression.
OLS regression
The OLS regression is expressed as the following:
Rit = a + b COVID-19 + s GOLDt + g GTit + p OILt + d S&P 500t + f VIXt + εit (2)
ability 2021, 13, x FOR PEER REVIEW 8 of 18
where i represents Rit = a + fiveb COVID-19 + s GOLD
cryptocurrencies (i t=+1gtoGT 5);it t+represents
p OILt + dthe S&P 500t +from
period f VIX22/01/2020
t + εit (2)
towhere
11/04/2020; Rit represents
iRrepresents
it = a + b COVID-19
daily
five cryptocurrencies cryptocurrency
+ s GOLD t +(ig=GT 1 to it +
returns;
p tOIL
5); a is the
t + d S&P
represents the intercept;
500 t + f VIX
period εitt is
from εthe
+ 22/01/2020
it
error
(2)
term; b, s, g, p, d,
to 11/04/2020; Ritand f are thedaily
represents sensitivity coefficients
cryptocurrency for theaindependent
returns; isthe theperiod
intercept; variables of the
where
regression. i represents five cryptocurrencies (i = 1 to 5); t represents fromε22/01/2020
it is the error
Sustainability 2021, R it = a + b COVID-19 + s GOLDt + gterm;

13, 8578 GTit +b,ps,OIL g, p,t +d,d and
S&Pf500 are + f VIX
t the t + εit
sensitivity (2)
coefficients
to 11/04/2020; Rit represents daily cryptocurrency returns; a is the intercept; εit is the error for the independent variables of the 8 of 17
where i represents five cryptocurrencies (i term; 1regression.
to b,
=Quantile 5);s,t represents
g, p, d, andthe
Regression f are the sensitivity
period from 22/01/2020 coefficients for the independent variables of the
to 11/04/2020; Rit represents daily cryptocurrency regression. returns; a is the intercept; εit is the error
Following
Quantile Regression[49], we deployed quantile regressions for conditional median-based
term; b, s, g, p, d, and f are the sensitivity coefficients for the independent variables of the
estimation. Following The loss [49],function
we deployedin statistics or mathematics
quantile regressions is presented
for conditional below.median-based
Sustainability 2021, 13, x FOR PEER REVIEW Quantile Regression
regression.
Ʈu if u>0
8 of 18
OLS Regression estimation.
Following The losswe
[49], function
deployed in statistics or regressions
ƍT={ quantile
mathematicsfor is presented
conditional below.
median-based
Quantile Regression Ʈu if u>0 (3)
Theestimation.
OLS regression The loss function in statistics
is expressed (Ʈ-1)uor asif mathematics
the following:
u≤0 is presented below.
Following [49], we deployed quantile regressions for conditionalƍTmedian-based ={
Ʈu if
= u (Ʈ- I(u<0)) u>0 (3)
estimation. The loss function in statistics or mathematics The Ʈth quartile is presented minimizes ƍT={ξ, as(Ʈ-1)u
below. shown
if u≤0
in the equation below (univariate):
= u (Ʈ- (3)
Sustainability 2021, 13,Ʈu
x FORRitifPEER
= itaREVIEW
u>0
R =+ ab+Ʈth COVID-19
b quartile
COVID-19 + s+GOLD s(Ʈ-1)u
GOLD +I(u<0))
ift u≤0 tg +in GT
gtheGT it + it + ppOIL OIL t+ t +ddS&P S&P500 500t t++8 foff VIX
VIX
18 tt +
+ εεitit (2)
(2)
ƍT={ The minimizes =ξ,Ruas shown
(ξ)=
(Ʈ- I(u<0)) ƍT(Yi −equation
𝜉) below (univariate): (4)
(3)
where (Ʈ-1)u if u≤0The Ʈth quartilecryptocurrencies
i represents minimizes ξ, asRshown (ξ)= (i in =
(i theƍT(Yiequation
1thattoit−5); 𝜉)ttabelow
represents (univariate): the period (4) from 22/01/2020
= u (Ʈ- I(u<0))
The special five cryptocurrencies
feature of the above function =is has represents
bidirectional derivative period but 22 January
to
2020 11/04/2020;
to 11 is not
April differentiable,
R
The represents
2020;
special Ras
feature showndaily below:
represents
of the R
abovecryptocurrency
(ξ)= daily
function ƍT(Yiis − 𝜉)
cryptocurrency
that it returns;
has a bidirectional a is the
returns; intercept;
derivative a (4)
is
butthe ε is the
intercept; error
18εit
Sustainability 2021, The Ʈth quartile minimizes ξ, as shown in the equation
Sustainability
13, x FOR 2021,
PEER 13, REVIEW
x FOR PEER REVIEW it Rit =below
a + it b COVID-19
(univariate): + s GOLDt + g GTit + p OILt + d S&P 500t + f VIXt + εit (2) it 8 of 8 of 1
Right
isg,not derivative as
differentiable, shown below:
term;
is the b,
R (ξ)=
s,
error ƍT(Yi Thep,
term;
where d,
specialand f
feature
−i derivative
𝜉)
represents are
b, s, g, five of the
the sensitivity
p, cryptocurrencies
d, and
above f are
function
R’(ξ+) = Lim (4)
coefficients
(ithe
is that
= 1 to(R(ξ
(h0+) sensitivity
it has a
5); t+represents for
bidirectional the
h)-R(ξ))/hcoefficients independent
derivative
the period from 22/01/2020 variables
for the independent
but of the
is notto Right
differentiable,
regression.
variables of the 11/04/2020;
regression. Ras shown below:
it represents daily cryptocurrency returns;
= Lim(h0+)
R’(ξ+) = Lim
ƍ (
(h0+) (R(ξ + h)-R(ξ))/h
a) isƍ (the )intercept; εit is the error
The special feature of the above function Right thatderivative
isterm; it s,
b, has g, ap,bidirectional
d, and f are the derivative
sensitivity butcoefficients for the independent variables (5)of the
ƍ ( ) ƍ ( )
isSustainability
not differentiable, as shown
2021, 13, x FOR below:
PEER REVIEW regression. = Lim(h0+)
R’(ξ+) = Lim(h0+) (R(ξ + h)-R(ξ))/h 8 of 18
= ∑ (𝐼(𝑦𝑖 ≤ 𝜉) − (Ʈ) ) ( ) (5)
Right derivative QuantileRRegression it = a + b COVID-19 Rit = a + b COVID-19 + s GOLD = Lim(h0+) t + g s GOLD
GTit ƍ+ p t +OIL g GT ƍ t +itd + S&P
p OIL 500 t +t d + S&P
f VIX500 t + ε t it+ f VIX(2)t + εit (2
R’(ξ+) = Lim Left derivative
Quantile
(h0+) (R(ξ +Regression
h)-R(ξ))/h = ∑ (𝐼(𝑦𝑖 ≤ 𝜉) − Ʈ) (5)
Following
Following [49], we deployed quantile regressions for conditional median-based esti-
where i represents
= Lim(h0+) where i[49],
Left derivative
ƍ (
five
represents
Following
we ) ƍ deployed
(
cryptocurrencies
[49], we
)
five R’(ξ-) quantile
= ∑ =quantile
cryptocurrencies
deployed (iLim = (h0+)
(𝐼(𝑦𝑖 1≤to𝜉) regressions
(R(ξ–-
5);
− Ʈ)
regressionst(ih)- =R(ξ))/h
represents
1for 5);for
to conditional t the conditional
represents
period
median-based themedian-based
from period
22/01/2020 from 22/01/202
Sustainability 2021,
Sustainability
13, x FOR 2021,
PEER 13,REVIEW mation.
x FOR The
PEERRitREVIEW
estimation. =a+b
The loss
COVID-19
loss
estimation. function + s GOLD
function
The loss in statistics
t + g GTit + p or
in
function statistics
in=
OIL(5)
Lim(h0+)
R’(ξ-) =
statistics mathematics
Lim
t + d S&P 500
or
or mathematics
(R(ξ–-
mathematics
ƍ (t + f VIX
h)- is
)isƍ tpresented
R(ξ))/h
(+ εit)
is
presented
(2)
presented
below. below. below. 8 of 18 8 of 1
to 11/04/2020; to 11/04/2020;
Left Rit represents
derivative Rit represents
daily cryptocurrency daily cryptocurrency
(h0+)
returns; a isreturns; the intercept; a is(6)theεintercept;
it is the error εit is the erro
where = ∑ (𝐼(𝑦𝑖 ≤ 𝜉)cryptocurrencies
− Ʈ)
term;
i represents
b, s, g, term;
five (i = 1 to Ʈu
5); tƮu
represents
= Lim(h0+)
R’(ξ-) = Lim if
if u>0
u>0 the ƍ ( from
period
(R(ξ–- h)-coefficients
R(ξ))/h
) ƍ22/01/2020
( )
Rp, d,b,and s, g,f are p, d,the and sensitivity

f are the Ʈcoefficients
sensitivity for the independent
for the independent variables ofvariables the of th
(h0+)
ƍ=T={returns; u−a 𝐼(𝑦𝑖
is the<intercept;
𝜉)

ƍ ( if u ) εƍ> it is 0
to 11/04/2020; it represents daily cryptocurrency
( the ) error
(6)
Left derivative
regression.
term; b, s, g, regression.
p, d, and f are the sensitivity ƍ
dT T ={ = =
coefficientsLim(h0+)for
(Ʈ-1)u theif independent
u≤0 variables of the
(3)
R’(ξ-) = Lim(h0+) Point (R(ξ–- h)- R(ξ))/hthe objective
ξ minimizes = function (Ʈ −−if𝐼(𝑦𝑖 )u< 𝜉)≥ if0 and
1R’(ξ+) u ≤R’(ξ-) 0 ≥ 0. Hence, (6) the (3)
(3)
regression. ƍ ( )defined
ƍ ( (Ʈ-1)u
) the unconditional= u (Ʈ-ifI(u<0)) u≤0
Roptimized
= Lim(h0+) it = aPoint bproblem
+ The COVID-19
ξ R = a + b COVID-19
+ s
Ʈth quartile minimizes ξ, as shown
minimizes
it the =
GOLD
objective= u ( Ʈ
function
t +−
− gs𝐼(𝑦𝑖
I (
GOLD
GT u
if <
quantile
< 𝜉)
R’(ξ+)+ 0 p ))above.
+
≥ OIL
0g GT
and
in the equation below (univariate):
it t t
InR’(ξ-)
+ itda similar
+ S&P
p ≥OIL
0. way, the
500
Hence,
t + t d+ S&P
f
theVIX 500
t + εt +
it f VIX(2) t + εit (2
the =unconditional
u (Ʈ-
conditional quantile can be defined analogously (6) I(u<0)) by OLS, asabove. explained
Quantile Regression
Quantile Regression Quantile
optimized
Point ξ
Regression
problem
minimizes defined
the objective function if quantile
R’(ξ+) ≥ 0 and In abelow.
R’(ξ-) ≥similar
0.
If [Y1,
Hence, way,the Y2,
the
ƍT(Yi
where = The
Following Ʈth
…,
i represents −Yn]
Ʈconditional
th
where
optimized
𝐼(𝑦𝑖
[49],
is a<set
quartile
quartile
iproblem
we five of random
𝜉)quantile
representsminimizes
deployedminimizes variables
cryptocurrencies
can
defined five
be
quantile ξ,
ξ, from
as
as R it,
cryptocurrencies
defined
the
(ξ)=
shown
shown
(i
analogously
regressions
unconditional
we get in
= for in
1 quantile the
the
toconditional
by5); OLS, t− =𝜉)1explained
equation
equation
(iabove.
represents
as toIn5);
median-based tbelow
below
the
represents
below.
a similar If(univariate):
period
way, (univariate):
[Y1, from
the
theY2,
(4)
period
22/01/2020 from 22/01/202
Following …, Yn] Following
[49],
is a set we
of deployed
random [49], we
variables quantile
deployed
from it, we getregressions
quantile regressions
for conditional for conditional
median-based median-based
Point ξ minimizes the objective estimation. The loss
to 11/04/2020;
to 11/04/2020;function
The
Rit represents
conditional
function ≥ 0inand
special
Rit represents
if R’(ξ+) statistics
daily cryptocurrency
quantile feature
daily cryptocurrency
can or 0.RHence,
returns; a isreturns;
be defined
R’(ξ-) (µ)= function
of≥mathematics
the above
the intercept;
a is theεintercept;isƍT(Yi
it is the error
analogously the presented
by is −that
εit is the erro
OLS, 𝜇) as below.
it explained
has a bidirectional below. Ifderivative
[Y1, Y2, (7) but
optimized problem definedestimation. estimation.
The loss function
TheƮu loss
asin function
ifstatistics
u>0 R(R ∑
n
)(ξ)=
it,=in
or
thestatisticsƍT(Yi
mathematics or− ξ𝜉)
−mathematics
is
) presented is below.
presented below. (4)
fR T𝜇)(Yi
…, Yn]
the unconditional is is
not differentiable,
a set
quantile of above.
random ashown
Invariables
similar below:
fromξ
way, we get
ƍT(Yi
term; b, s, g, term;
p,Thed,b, and
s, g,f are
p, d,theand sensitivity
are (µ)=
the i =1 −from
is coefficients
sensitivity coefficients
for the independent
for(7).
the (7)
independent
variables ofvariables
of th 8 the
unconditional population mean estimated the above Equation Then,
Sustainability 2021, 13,quantile
conditional x FOR PEER defined analogously byRight
can be REVIEW asƍexplained
T={
OLS, derivative below.Ʈu If [Y1, if Y2,u>0 Ʈu if u>0 of 18
=ƍT(Yi − 𝜇)from
the parametric function µ (x, β) replaces
R (µ)= the scalar µ in the above equation to
(3) get Equation
(7)
regression.
…, Yn] is a set of random variables The
from regression.
it,special
we get The unconditional
feature of population
the
(Ʈ-1)u if above
u≤0 meanR’(ξ+) is estimated
function is
Lim(h0+) that
(R(ξ +the
itabove
has
h)-R(ξ))/hEquation
a (7). Then,
bidirectional derivative but
The (8).
special
theThe feature
parametric of
function ƍ(x,T={
uthe
µ(Ʈ- above
β) replaces ƍT={
function scalar µisfrom
inthat ƍ it
( has ) a
ƍ (bidirectional
)to get Equationderivative but is
is not differentiable, unconditional
as
= population
shown
I(u<0))mean=isthe
below: Lim(h0+)
estimated thethe
above equation
above Equation (7). Then, (3) (3)
RThe
(µ)=
Ʈth ƍT(Yi
(8). − 𝜇) (Ʈ-1)u (7)
if u≤0
ƍT(Yi (Ʈ-1)u
− if u≤0
not differentiable,
thequartile
parametric as
minimizesshown
ξ, as
function µ (x, below:
shown R in
(µ)
β) replaces the
= equation
the scalar in𝜇 the
below
µ (x, β))
(univariate):
above equation to get Equation (8) (5)
Quantile RightRegression
Quantile
derivative Regression
The unconditional population mean Right (8).estimated
is derivative from the
Sustainability
Similarly, the2021,
above
R (ξ)=
13,
conditional
Equation
x FOR = ƍT(Yi
R
PEER (µ)
median u
REVIEW= (Ʈ-
(7). =𝜉)
−Then,
function ∑I(u<0))
ƍT(Yi =(𝐼(𝑦𝑖
can −ube 𝜇≤(Ʈ- 𝜉)
(x, −I(u<0))
β))
obtained Ʈ) by replacing (4) the scalar (8) 8 of 1
the parametric function µ (x, β) replaces
The
theRscalar
Following
Ʈth it =Lefta µ[49],
quartile
variable The+
ξ
inb theCOVID-19
Following
with Ʈth
above
we
minimizes
the
derivative
equation
deployed
quartile[49],
parametric + to
wes get
Rminimizes
ξ, as
function
(µ) R’(ξ+)
GOLD Equation
quantile
deployed
=thatshown
ξ (x t
ƍT(Yi
t, =
β) + Lim
g
ξ,
and in GT
regressions
as quantile
(h0+)
𝜇the
−bidirectional
by (x,it
shown + p(R(ξ
equation
β))
setting OIL
the inƮth+ t
the+h)-R(ξ))/h
regressions
for d S&P
conditional
below
quantileequationas½2. 500 for t
(univariate):
Other +
below
(8) f median-based
VIX(univariate):
conditional t + εit (2)
median-based
(8). The special Similarly,
feature of the the conditional
above function median
is function
it has a can be obtained derivativeby replacing
but the scalar
estimation. conditional
estimation.
The loss function
functionThe 0 loss
values incan beLim
function
statisticsobtained =by replacing absolute −(values
) ƍis ) /h
with ƍ is ƍ( T(*), )as
(ξhin or 0statistics
mathematics or mathematics presented below.
presented below.
Sustainability 2021, 13, x FOR PEER REVIEW (ξ+)
Rparametric = R’(ξ-) Lim (h0+) (R(ξ–-
(toRby + hh)- R(ξ))/h Rquantile 8 of 18
variable
is not differentiable,
where asξ shown
with thethe below: function= function
Lim(h0+)
(x
→ t, β)
+) and be(5); ξsetting the Ʈth (ξ)) as½2. Other
= i represents five cryptocurrencies (i = 1 t represents the period from 22/01/2020
Similarly, conditional median can obtained ƍ ( by ) replacing the scalar
R (µ)Right ƍT(Yi
follows:
conditional
derivative
variable
− 𝜇 (x, β))
ξ withfunction
the parametric can Ʈu
values function R (ξ)=
be ξ=
obtained t, if
(8)
(xLim(h0+)
β) andnRby
u>0 =R
it Ʈu
ƍT(Yi
+(ξ)=
replacing
byaT −ξif
(byCOVID-19
setting −− h𝜉)
absolute
u>0
the Ʈth
)− ƍT(Yi
+T
ƍ (
−ξ−
s(values
yGOLD
quantile
)
𝜉) Other
with
) as½2.
t + g GT
ƍT(*), as
it + p OILt + d S&P 500t + f VIXt + εit
(4)(5) (4
(2
to 11/04/2020; follows: Rit represents =values
Lim =(h
daily → 0+)
cryptocurrency ∑ returns; a is the intercept; (6)εit is the error (5)
R’(ξ+) Lim
Rbe (ξ) =the (R(ξ ƍT(Yi
+by 1− 𝜉 (x, β))
h)-R(ξ))/h
i =replacing h values with ƍT(*), as (9)
ƍ ƍ
(h0+)
Similarly, the conditional median function conditional can be function
obtained can
bynreplacing
T={
obtained scalar ={ absolute
variable ξ with the parametricterm; The
function special
b,ξ (x
s,follows:
t,g, p,
β) and Thefeature
d,
by and special
setting =
ofƮth
fthe
are ∑
=the
feature
the=1above
Lim(h0+)
quantile I (Ryi ofwhere
(sensitivity
as½2. ≤ = ξ∑
=Other
function
the T
− ( Ʈ(𝐼(𝑦𝑖
)iƍabove
represents )−−)is𝐼(𝑦𝑖
coefficients
ƍT(Yi
≤
( five
ƍfunction
that <) 𝜉) 𝜉) −
β))itfor has Ʈ)athat
cryptocurrencies
is
the it(i =has
bidirectional
independent
1 to 5); t represents the period from 22/01/202
a by bidirectional
derivativeof
(9) variables but
derivative
(3) the εit is the errobu
(3)
Further, using linear iprogramming, (ξ)to=11/04/2020;
(Ʈ-1)u theifminimizing
u≤0 Rit 𝜉represents
(Ʈ-1)u
(x,problem
if u≤0 can
daily easily (5)be solved returns;
cryptocurrency a is the intercept;
conditional function valuesis can
notbe obtained
differentiable,
is by
not replacing
differentiable,
as absolute
shown values
below:
as with
shown ƍ T(*), as
below:
regression.Sustainability
Left derivative formulating
Rit =used a +optimized
2021,
Point
b COVID-19 ξ ξ13,
(x, β) x FOR
minimizes
as linearPEER
+defined
the
s GOLD
REVIEW
objective
parameters.
term;
R (ξ) = t + the b,function
ƍT(Yi
The s, g, − 𝜉 (x,
authors
p, d, if andR’(ξ+)
β))
of f ≥
[62–65]
are 0
the and
and R’(ξ-)
others
sensitivity ≥ 0. Hence,
extensibly
coefficients (9) the
for the independent variables of th
follows: Left Further,
derivative
this method
using linear
problem
recently = programming,
∑
to derive
(𝐼(𝑦𝑖
= the ≤
the 𝜉) −gI(u<0))
uunconditional
(Ʈ-
study Ʈ)GT = it +uquantile
minimizing
estimates.
p(Ʈ- OIL problemt + dcan
I(u<0))above.S&P In a 500
easily be tsolved
similar + fway, VIXby t + εit
the (2)
Right derivative
formulatingRight ξ derivative
(x, β) as linear regression.
parameters. The authors of [62–65] and others extensibly
where Left
R
iThe
(ξ) = Ʈth
derivative
represents ƍT(Yi
used
Further,
conditional
quartile
−The (x,Ʈth
using
𝜉 method
five
this β))
linear programming,
quantile
minimizes
quartile
cryptocurrencies
can be asR’(ξ-)
ξ,defined
minimizes shownthe
(i(R(ξ–-
==The
= ξ,
analogously
1(9) to
Lim
minimizing
in as
5); the byproblem
(h0+)
tshown
OLS, (R(ξ–-
equation
represents
as
in can
the h)-
explainedeasily
below
the
R(ξ))/h
be
equation
period
solved
below. If by[Y1, Y2, (univariate):
+(univariate):
below
from 22/01/2020
Sustainability 2021, 13, x FOR PEER REVIEW Quantile Regression
formulating
…, Yn] ξis (x,
a set β)recently
of linearto
as0random
R’(ξ-) =
deriveR’(ξ+)
parameters.
variables
Lim
the
Quantile
study
from Lim estimates.
R’(ξ+)
authors
it, we
Regression
h)- (h0+)
get
R(ξ))/h =[62–65]
of(R(ξ Lim +ƍ(h0+)
h)-R(ξ))/h
and ( (R(ξ
others ) h)-R(ξ))/h
extensibly
ƍ ( ) 8 of 18
Further, using linearto
programming, used
the minimizing
R (can ξ−) = be Lim =
(h0+)
Lim(h0+)
(h→ +) ()R
0estimates. (ξ − h) − R(ξ))/h
11/04/2020; R itthis method
represents recently
problem daily toeasily
derive the
cryptocurrencysolvedstudy by returns; ƍa is
( the )intercept;
ƍ (
ƍ ( ) ε ) ƍis (the ) error(4)(6)median-based
Rextensibly
= (ξ)= ƍ (
RƍT(Yi ƍ
(ξ)= ( )
− 𝜉)
)−ƍT(Yi ξ− ) 𝜉) regressions
it
0Lim(h0+) n =
Following [49], we =deployed Lim(h0+) quantile
R (µ)=Following TLim(h0+)
[49],
regressions
ƍT(Yi (y− −ξ we +hdeployed
𝜇) Tfor (y−quantile
conditional for conditional
median-based
(7) (4
p(6)
formulating ξ (x, β) as linear parameters. The authors of [62–65] and others
term; b, s, g, p, d, and f are =the Lim ( h→
sensitivity +) ∑
coefficients
estimation. i = The
1 R it =
loss a for+ b
function the
hCOVID-19
inindependent
statistics (6) or+mathematics
s GOLD variables t +
is g GT
of
presented it +below.
the (5) OILt + d (5 S
estimation. The loss
used this method recently to derive the study estimates. function
The unconditional =∑ n in statistics
population
Ʈ− =
𝐼(𝑦𝑖 mean
< 𝜉)
oris Ʈ mathematics
−
estimated 𝐼(𝑦𝑖 <
from theis
𝜉) presented
above Ʈu Equation if u>0 (7). below.Then,
regression. The special Thefeature special =
of the
feature (
above
i =1µ (x, β)Ʈu− ofI ( yi
function
the < above
ξ )) is function
that it has is athatbidirectional
it has a bidirectional
derivative but
derivative bu
the parametric function = ∑ if
replaces
where i(𝐼(𝑦𝑖
the u>0 scalar
represents =≤ µ∑ 𝜉)the−
in (𝐼(𝑦𝑖
above
five Ʈ)ƍT={ ≤ 𝜉) −toƮ)
equation
cryptocurrencies get Equation
(i = 1 to 5); t represents(3)t
is notPointdifferentiable,
Rit ξ=isminimizes
aξnot
+(8).
b differentiable,
COVID-19as
the shown
objective + s below:
as
GOLD
function shown t + gbelow:
if R’(ξ+) GT
≥ 0 it and + p OIL
R’(ξ-) ≥t 0.+ d
Hence, S&P the
(Ʈ-1)u 500 if t + f VIXt + εit
u≤0 (2)opti-
optimizedPoint ξ
problem
minimizes
minimizes thethe ƍ
objective
objective
T ={ to function
defined the unconditional quantile above. In a similar way, function
11/04/2020; if R’(ξ+)
if
R R’(ξ+)
it ≥
represents 0 ≥ and 0
= uthe
andR’(ξ-)
daily R’(ξ-) ≥ 0.
cryptocurrency≥Hence,
0. Hence, the the
returns; a is th
Left
Quantile derivative
Right
Regression LeftRight
derivative derivative derivative R (µ) = ƍT(Yi − 𝜇 (x, β)) (Ʈ- I(u<0)) (8) (3)
where mized
optimized
conditional problem
i represents problem
quantile defined
five
can be defined theanalogously
defined
cryptocurrencies unconditional
the unconditional
(Ʈ-1)u
term;
by(iOLS,=The 1b, quantile
ifto
as
Ʈth u≤0
s, 5);g,
explained
quartile tp, above.
quantile
represents
d, and
below.
minimizes f
If In
arethe
[Y1, aY2,
ξ,above.
as similar
period
the
shown In
sensitivity
in the way,
a equation
from similarthe conditional
22/01/2020 way,
coefficients
below the for
(univariate): the ind
…, Yn] is a set
Following
quantile can of[49],
random
be variables
we
defined
Similarly, deployed
the from
analogously
conditional
R’(ξ-)
R’(ξ+)
it, wequantile
get
median by = OLS,Lim R’(ξ-)
R’(ξ+)
(h0+)
regressions
function as can
(R(ξ–-
=beLim
(R(ξ
explained +for
obtained
h)-conditional
h)-R(ξ))/h
(h0+) R(ξ))/h
(R(ξ–-
(R(ξ
below.
by replacing
h)-
+Ifh)-R(ξ))/h R(ξ))/h
the median-based
[Y1, Y2,
scalar . . . , Yn] is a
conditional quantile
to 11/04/2020; Rit represents daily cryptocurrency can be defined = uanalogously
regression. (Ʈ- I(u<0)) by
returns; a (is the OLS, as explained
intercept;
R quantile
(ξ)= ƍT(Yi below.
ε−)it 𝜉) If
is the) error [Y1, Y2, (4
) ƍ ƍ( (as½2. ) Other ƍ (
estimation.
term; setb,Yn]
…, ofThe The
s, random
g,is Ʈth
p,a d,set
variable ξ with theRparametric
loss offunction
variables
quartile
and
conditional random
f are from
minimizes
the
function
in(µ)= statistics
variablesit, we
valuesξ,
sensitivity =
function
ƍT(Yi
canas
Lim(h0+)
get
from − 𝜇)
or
beshown
ξ (xt, β) and by setting
mathematics
coefficients
obtained it, we=
inby
Lim(h0+)
get
the for
replacing
isƍabsolute
equation
the
the Ʈth
presented
independent (7) below.
below
values (univariate):
with variables
ƍ (*), as of the
The special feature of the above function is that it has a bidirectional T
(5) derivative (6
(6) (5
bu
regression. The unconditional follows:population mean isƮu Quantile
estimated
is not if from
u>0the
differentiable, Regression
aboveas Equation
shown below: (7). Then,
the parametric function µ (x, β) replaces ƍT={the scalar R
R=(Rµ=µ(ξ) (µ)=
)(ξ)=
∑
in= the Ʈ
Right ∑above
n
(𝐼(𝑦𝑖− =1 𝐼(𝑦𝑖
derivative ƍT(Yi
= ≤(∑𝜉Yi
equation
T− 𝜉) <− − −
Ʈ(𝐼(𝑦𝑖
𝜉)µ
to − 𝜇)
𝜉)
Ʈ)
2
) 𝐼(𝑦𝑖
get ≤ 𝜉)
Equation <−𝜉)Ʈ) (7)
(4)
(7)
= Following iƍT(Yi
= (x, β))
[49], we R’(ξ+) deployed (9)
(8). = Lim(h0+) quantile (R(ξ + h)-R(ξ))/h regressions
(3) for c
Quantile Regression
The
The unconditional
special feature population
of the (Ʈ-1)u
above mean if u≤0
function is estimated
is that it from
has athe above
bidirectional Equation derivative (7).
) ƍ ( Then, )but
LeftPointderivative Left
ξ minimizes
The unconditional derivative
Further,
Point ξ minimizes using
R the linear
objective
population
(µ) = ƍT(Yi estimation.
programming,
the −
mean function
𝜇 objective
(x, the
β))isThe The
minimizing
estimatedif function loss
R’(ξ+)fromproblemfunction
≥ if0 =the can
R’(ξ+)
and in
easily
Lim(h0+)
(8)above statistics
be solved
R’(ξ-)
≥ 0Equation and ƍ or
by
≥ 0. R’(ξ-)( mathematics
Hence, (7).≥Then, 0. Hence,pres
the is th(5
formulating ξ (x, β) as linear =below: uparameters.
(Ʈ- I(u<0)) authors of [62–65] and others extensibly
the
is not
optimized
the
parametric
Followingdifferentiable,
parametric [49],
optimized
problem function
we as
defined
function problem µ
shown
deployed (x,
µ
β)
the
(x,
replaces
quantile
R’(ξ-)
defined
unconditional
toβ) replaces =
bethe
the
Lim scalar
regressions
R’(ξ-)
unconditional
(h0+)
the =µ
quantile
scalar
in
Lim
(R(ξ–-
replacingµthe
the
for h)-
(h0+) above
conditional
above. R(ξ))/h
(R(ξ–-
quantile
in the
equation
In h)-
above a Ʈu
above.similar to
median-based
R(ξ))/h if
In
equation
getu>0
way,
a Equation
similar
to the get way, th
TheRight
(8).
estimation. ƮthThe
Similarly, the
quartileusedconditional
derivative
loss
this method
minimizes
function
medianrecently function
ξ, as shown can
derive the obtained
study
in the equation by
estimates.
below
( asby
scalar
)OLS,
(𝐼(𝑦𝑖 ≤ 𝜉) − Ʈ)
= ∑(univariate):
ƍƍƍ(Tbelow.
( as ) below. ) ƍ ( If )[Y1,
variable
conditional
Equation ξ with the
conditional
quantile
(8). parametric can beindefined
function
quantile statistics
ξ (xcant, β) and by
= be or
analogously
defined
Lim(h0+)
mathematics
setting
=
Ʈth
theanalogouslyby
Lim(h0+)
quantile
OLS, isƍ as½2.
presented Other
explained ={ explained below. Y2, If [Y1, Y2
Left derivative ƍT(*),
conditional function values can be obtained
…,follows:
Yn] is a…, set Yn] of randomis a set of variables
random ƮuR
byR’(ξ+)
(ξ)=
from
replacing
if
= =∑i=1 ƍT(Yi
variables u>0
n =
it,
absolute
Lim
ƍT(Yi
we (h0+)
from
get
T(Yi −−µ𝜇(x, −
values
(R(ξ
it,𝜉) we
with+ geth)-R(ξ))/h
2 as
(Ʈ-1)u if u≤0 (4) (6) (6
R (R µ)(µ) (x,β β)) )) R’(ξ-) = Lim (h0+) (R(ξ–- h)- R(ξ))/h (8)
ƍT={ == Lim(h0+) Ʈis − =ƍT(Yi 𝐼(𝑦𝑖it− < Ʈ𝜉)
ƍ (
−a𝐼(𝑦𝑖 <
) ƍ ( = ) u (Ʈ- ƍ ( I(u<0)) ) ƍ ( )
(9) 𝜉)
= Lim(h0+) (3)scalar
TheSimilarly,
special feature
Similarly, the ofRthe
the conditional
conditional (ξ) =above ƍT(Yi
median
(Ʈ-1)u
median
− 𝜉 (x,function
Rfunction
(µ)= if Theu≤0
β))
function R
Ʈth (µ)=
that canhas
quartile
can 𝜇)
beƍT(Yi
be obtained
minimizes
obtained −
bidirectional 𝜇)by by asderivative
ξ, replacing shown the
replacing in but
the the (7)(5)
equation
scalar (7
be(6
is not differentiable, as shown below: = Ʈ − 𝐼(𝑦𝑖 < 𝜉) 1
variableFurther, ξ with
using
Pointunconditional
ξξξwith the
linear
minimizes
Point parametric
programming,
ξ population
minimizes
the objective thefunction
minimizing
the ξ (x
problem , β) and
can by
easily setting
be solved the
by
Ʈth th quantile as 2. Other
variable
The
formulating (x, β)The
the parametric
unconditional
as linear
=function
parameters. Thepopulation
umean(Ʈ- =function
authors
objective
I(u<0))
∑ξis tt(𝐼(𝑦𝑖
(x[62–65]
, β)
of estimated mean if
and function
≤
and
R’(ξ+)
by
𝜉)
is −
from ≥Ʈ)ifthe
setting
estimated
others
0 R’(ξ+)
extensibly
and
the
abovefromR’(ξ-)
≥ Equation
0 and
quantile
the ≥above0. R’(ξ-)
Hence,
as½2.
(7). ≥Other
2 Then,
Equation
ƍT(Yi
the
0. Hence,
− (7).
𝜉) Thenth
Right
conditional
optimized
The
conditionalƮthderivative
optimized
problem
quartile function
function defined
minimizes values
problem
values can
the
ξ,
can defined
as be
shown
be obtained
unconditional
obtainedPoint
the in the by
ξunconditional
minimizes
by replacing
quantile
equation
replacingthe objective above.
below absolute
quantile
absolutefunction
In
(univariate): a ifRsimilar
values
above.
values
(ξ)=
R’(ξ+) with
≥ 0 way,
In
with and
a ƍ (*),
R’(ξ-)
similar
T the
(*), ≥as
as0.way,
Hence,th th
theused this method
parametric recently
the parametric
function to deriveµ (x, the
functionβ)study
replaces estimates.
µ optimized
(x, β) thereplaces scalar µthe in thescalar above µ inequation
the above toequation
get Equation T
Left
follows: derivative R’(ξ+) = Lim problem (R(ξ defined
+ the
h)-R(ξ))/h unconditional quantile above. Into a get
similarEquation
way, th
conditional
follows: (8).
(8). conditional
quantile can quantile
be defined can be
analogously
defined
The
n ƍT(Yi
(h0+) analogously
special by OLS,
feature as byexplained
OLS,
ofanalogously
the above as below.
explained
by function If [Y1, below.
is thatY2, If [Y1,
it has Y2
a Y2
bi
R (Rξ)(ξ)= =
conditional
R’(ξ-) ∑ =is1Lim
quantile
Tfrom
(ofYirandom
(h0+) − can
−it, ξ𝜉)
be defined
(R(ξ–- (x, βh)-
ƍ (variables )))2fromR(ξ))/h
ƍ ( it, we ) get
OLS, as explained
(4) below.
(9)
If [Y1,
…, Yn] is a…, set Yn] of randomis a set of variables
random from
= Lim(h0+)
…,
is variables
Yn]it,
not i= awe set
differentiable, get we asget shown below:
RR (ξ)= = RƍT(Yi ƍT(Yi = −−𝜇 𝜉(x, (x,
ƍT(Yi ƍ β))
β)) (
− 𝜇) (x, ƍ (
β)) ) (5) (9)
The Further,
special feature of theprogramming,
above (µ) = Lim(h0+) (µ) ƍT(Yiderivative
− 𝜇) be solved but(8)by (8
using linear Rfunction
(µ)= Right the isRƍT(Yithat
(µ)=it−has
derivative
minimizing a problem
bidirectional
− 𝜇)can easily
R (µ)= (7
𝜇) ƍT(Yi (7)(6) (7
is not differentiable,
formulating
Further, ξusing (x,as β)shown as linear
linear below:
programming, =
parameters. ∑ (𝐼(𝑦𝑖
The the The ≤ 𝜉)
minimizing
unconditional −
authors Ʈ)
population of problem[63–65] can
mean is estimated and R’(ξ+) others
easily from=by theextensibly
Lim
be solved
above (R(ξ
bythe
(h0+)Equation +(7).
h)-R(
Then
Similarly, the
Similarly,
conditional the conditional
median function median can function
be obtained can be by obtained
replacing the
replacing
scalar scala
Right
used The
formulating unconditional
derivative
this method ξ The
(x, unconditional
recently
β) as population
linear to derive =thethe
population
mean
parameters. parametric
is
study ƮThe −
estimatedmean 𝐼(𝑦𝑖
function
estimates.
authors is<
µfrom 𝜉)β)ofreplaces
(x,
estimated the above
[62–65] the
from scalar
and µ inabove
Equation
the the above
others (7). equation
Equation
Then,
extensibly to (7).
get
ƍ Equatio
( Then
Left derivative
variable ξ with
variable the parametric
ξ with the parametric function (8). ξ (x function
t, β) andξby (xtsetting
, β) andthe by Ʈth setting =the
quantile Ʈth
Lim(h0+) as½2. quantile Otheras½2. Othe
the
usedparametric
this the method parametric
function recently µ (x,function
toβ)derive
replaces
beµobtained
R’(ξ+) (x, β)
== can
the the
Lim replaces
study scalar µthe
(R(ξ
estimates. in+thescalar byabove
h)-R(ξ))/h µ inequation
thevalues above toequation
getvaluesEquation
ƍ towith
get Equation
the ƍT(*), (8
(h0+)
conditionalPointconditional
function
ξ minimizes values
function
thecan R’(ξ-)
values
objective Lim
function be
(h0+) by obtained(R(ξ–-
replacing
if R’(ξ+) h)- ≥RR(ξ))/h
absolute
replacing
0
(µ) = and R’(ξ-) absolute
ƍT(Yi − 𝜇 (x, β)) ≥ with
0. Hence, T (*), as a
(8). (8). ƍ ( ) ƍ ( )
follows: follows:
optimized problem defined the = Lim(h0+)
unconditional ƍ (
quantile median )
above. ƍ ( )
In acansimilar = ∑ (𝐼(𝑦𝑖
way, ≤ 𝜉)
the − Ʈ)
= Lim(h0+) Similarly, the conditional function be obtained by(5)replacing the scala
conditional quantile can be defined R (µ) = analogously
variable R ξ ƍT(Yi
(µ)
with =
the − by 𝜇
parametric (x,
ƍT(Yi
OLS, β)) as
function− 𝜇 ξ(x,
explained(xt, β))
β) and by below.setting the If Ʈth(6)
[Y1, Y2,
quantile
(8) as½2. Othe (8
R (ξ) =∑Left derivative
= conditional RƍT(Yi
(𝐼(𝑦𝑖 (ξ) ≤=get
function −−
𝜉) 𝜉 (x,
values ƍT(Yi
Ʈ) β))
can be − obtained
𝜉 (x, β)) by replacing absolute values (9) with ƍT(*),(9 a
…, Yn] is a set of random variables = follows:from it,
Ʈ − 𝐼(𝑦𝑖 < 𝜉)we R’(ξ-) =byLim (h0+) (R(ξ–- h)- R
Similarly, the Similarly,
conditional the conditional
median function mediancan function
be obtained can beby obtained
replacing the
replacing
scalar the scala
4. Results and Discussion

4.1. Preliminary Analysis
Tables 1 and 2 show descriptive statistics and correlation estimates for the independent
variables. Most of the data series have higher levels of Kurtosis and correlation coefficients
within the acceptable range. Looking at the independent variables’ descriptive statistics
separately and their standard deviations as a measure of the absolute risk level of the assets,
it is obvious that gold, as a typical global commodity, has the lowest risk compared to all
other assets. The correlation analysis, with negative coefficients between gold and oil and
S&P 500 is a sign that these two global commodities could serve well in this COVID-19
crisis as hedge instruments against risky assets represented by the most famous stock
exchange index. The ADF test confirms that there is no unit root in the series.
Table 1. Descriptive statistics.
COVID-19 GOLD GT * OIL S&P 500 VIX

Mean 0.279 0.002 38.531 −0.002 −0.003 0.025
Std. Dev. 1.105 0.019 11.943 0.093 0.038 0.141
Skewness 2.911 0.451 2.526 0.242 0.328 1.098
Kurtosis 11.559 6.118 11.271 5.510 4.708 4.640
* Note: Google trend for Bitcoin.
Table 2. Correlation analysis.
COVID-19 GOLD GT * OIL S&P 500 VIX

COVID-19 1.000
GOLD −0.011 1.000
GT 0.029 −0.233 1.000
OIL −0.057 0.107 0.083 1.000
S&P 500 0.096 −0.095 0.110 −0.191 1.000
VIX 0.058 0.091 −0.147 0.009 −0.692 1.000
4.2. Wavelet Coherence

Figure 2 shows the wavelet coherence diagram for different cryptocurrency pairs. The
horizontal axis represents a daily timeline for one year, starting in May 2019. Our study
considers the COVID-19 period to be from 22 January 2020 and onwards in the horizontal
period axis (labeled 250 and onwards). The vertical axis represents scales at 2, 8, 16, 32, 64,
and onwards. The right-hand side vertical axis is the color axis: the brighter the color (red),
the more the coherence, and the lesser bright the color (blue), the lesser the coherence. The
Bitcoin–Ethereum, Bitcoin–XRP, Bitcoin–Bitcoin Cash, Ethereum–XRP, Ethereum–Bitcoin
Cash, and XRP–Bitcoin Cash pairs show very high levels of persistent coherence at all
scales during the COVID-19 period as compared to the pre-COVID-19 period. The wavelet
coherence diagrams of Tether with the other cryptocurrencies show interesting results,
especially during the COVID-19 period. During the COVID-19 period, especially on the
scale at around 16, Tether shows low levels of coherence with the other cryptocurrencies,
with no contagion effect. The direction of the phase arrows is not consistent across the
scales and periods.
Figure 2. Wavelet coherence diagram among the cryptocurrency pairs.

During the COVID-19 period, all pairs show a strong level of coherence with a short
contagion effect that starts around mid-February 2020 and lasts until April 2020 (Figure 2).
Such behavior supports the findings of [21], which show that Tether had a sizable impact
on Bitcoin prices. However, their findings were from before the COVID-19 period (we
found low levels of coherence for all cryptocurrencies before the COVID-19 pandemic),
therefore, our results have similar characteristics only during the pandemic. During the
COVID-19 period, at a scale of 64 and above, Tether shows lower coherence with the
other cryptocurrencies as compared to before the COVID-19 period. In addition, the
contagion effects that existed before the COVID-19 period at a higher scale of 64 and above
disappeared during the COVID-19 period. The above estimates highlight two important
things. First, during the COVID-19 period, the estimates are different, and secondly, they
are not uniform. The wavelet coherence diagrams for different cryptocurrency pairs have
clearly presented that all pairs, without the presence of Tether, show a very high level of
persistent coherence at all scales during the COVID-19 period. The behavior of Tether
regarding all the other cryptocurrencies in the crypto pairs is very specific, as these are the
only pairs with a low level of coherence with no contagion effect. Going further, during
the COVID-19 period, Tether shows lower coherence with the other cryptocurrencies as
compared to before the COVID-19 period. Therefore, it is obvious that Tether is a typical
stable coin and could present a safe haven for the entire crypto market because all other
cryptocurrencies used in the analysis present more than 80% of the total crypto market
capitalization. According to this, Tether is the only stable currency, while Bitcoin, Ethereum,
Bitcoin Cash, and XRP were affected by COVID-19’s influence. In the case of Bitcoin, this
study shows a similar behavior of this cryptocurrency as previously analyzed by [16].
Moreover, our results support more findings of [10,44], in which there is no evidence that
Bitcoin serves as a safe haven for global assets.
4.3. OLS Estimates

Table 3 shows the results obtained from the OLS estimates.
Table 3. OLS regression estimates.
C COVID-19 GOLD GT OIL S&P 500 VIX R2

Bitcoin 0.029 0.005 0.017 −0.001 0.116 ** 0.570 * −0.068 0.260
Ethereum 0.005 0.006 0.174 0.000 0.190 * 0.822 * −0.057 0.274
XRP −0.009 0.008 −0.033 0.000 0.112 ** 0.673 * −0.034 0.274
Tether −0.004 −0.002 * 0.036 0.000 −0.027 * −0.076 * 0.005 0.218
Bitcoin Cash 0.004 0.003 −0.178 0.000 0.156 ** 0.790 * −0.044 0.210
Note: * Significant @ 5% & ** Significant @ 10%.
The OLS estimates show that for all five cryptocurrencies, oil, and the S&P 500 index,
the variables are statistically significant at the 5% and 10% levels in explaining average
returns, as shown in Table 3. The COVID-19 variable is statistically significant only for
Tether at the 5% level in explaining average returns. Looking at the OLS estimates, it seems
that risky assets measured by S&P 500 have statistically significant influence at the 5%
level in explaining average returns of all five cryptocurrencies used in the analysis. As
one could expect, there is a positive change to S&P 500 on average returns for all other
cryptocurrencies except Tether (in the range of 0.57 for Bitcoin to 0.822 for Ethereum), which
is a clear sign that all these assets are typical examples of risky assets. The only reaction of
Tether to the change of S&P 500 is opposite to the other cryptocurrencies. This coefficient is
slightly negative (−0.076), which is one more piece of evidence for a typical stable coin that
is not directly influenced by the strong reaction of risky assets. The OLS estimates show
that apart from S&P 500, only oil, as a typical global commodity, has statistically significant
influence at the 5% and 10% levels on the average returns of all five cryptocurrencies. The
explanation for this is very similar to the previous one; the only reaction of the price of
Tether was negative but close to zero (−0.002). The only difference is that this influence
is much lower than for the case of S&P 500, with the other four cryptocurrencies in the
range of 0.112 for XRM to 0.19 for Bitcoin, which shows that oil still can be a diversifier.
The COVID-19 crisis seems to be statistically significant only for Tether, once again with a
negative reaction very close to zero (−0.002).
4.4. Quantile Regression Estimates

Table 4 shows the quantile regression coefficients of the right-hand side variables for
different quantiles, along with their associated p-values.
Table 4. Quantile regression estimates.
Coefficient Values
C COVID-19 GOLD GT OIL S&P 500 VIX
0.05 0.145 −0.001 −0.581 −0.006 0.295 1.696 0.166 Bitcoin
0.25 0.015 0.001 −0.151 −0.001 0.180 0.652 0.008
0.5 0.016 −0.002 0.008 0.000 0.123 0.620 0.022
0.75 −0.019 0.001 0.067 0.001 0.085 0.517 0.019
0.95 −0.076 −0.008 0.358 0.004 −0.145 0.487 −0.003
0.05 −0.092 0.024 1.630 −0.001 0.224 −0.221 −0.533 Ethereum
0.25 0.007 0.000 −0.024 −0.001 0.194 0.718 0.021
0.5 0.001 −0.001 −0.188 0.000 0.184 0.862 0.052
0.75 0.013 −0.007 0.321 0.001 0.114 1.207 0.062
0.95 0.045 0.002 0.680 0.001 −0.072 0.444 −0.158
0.05 −0.022 0.016 1.325 −0.002 0.162 0.192 −0.214 XRP
0.25 0.002 0.004 −0.350 −0.001 0.204 0.826 0.062
0.5 −0.028 0.003 −0.170 0.001 0.130 0.572 −0.008
0.75 0.004 0.003 −0.361 0.001 0.090 0.760 0.064
0.95 0.032 0.006 0.118 0.001 −0.012 0.432 −0.041
0.05 0.015 0.000 0.183 0.000 −0.035 −0.143 −0.007 Tether
0.25 0.002 0.000 0.036 0.000 −0.010 0.007 0.002
0.5 0.000 0.000 0.011 0.000 −0.003 0.005 0.000
0.75 −0.002 0.000 0.012 0.000 −0.009 −0.018 0.000
0.95 −0.002 −0.003 −0.193 0.000 −0.019 0.009 0.058
0.05 −0.164 0.022 0.934 0.001 0.306 −0.040 −0.460 Bitcoin Cash
0.25 −0.011 0.004 −0.235 0.000 0.238 0.750 −0.050
0.5 0.012 −0.001 −0.405 0.000 0.177 0.755 0.028
0.75 0.034 −0.006 0.118 0.000 0.135 0.734 0.012
0.95 0.153 −0.019 −0.722 −0.001 −0.210 0.810 −0.166
p values
C COVID-19 GOLD GT OIL S&P 500 VIX
0.05 0.000 0.802 0.247 0.000 0.026 0.000 0.044 Bitcoin
0.25 0.745 0.763 0.641 0.568 0.034 0.000 0.851
0.5 0.495 0.392 0.963 0.536 0.002 0.000 0.507
0.75 0.354 0.854 0.840 0.091 0.354 0.075 0.730
Table 4. Cont.
0.95 0.316 0.065 0.595 0.076 0.293 0.375 0.966

0.05 0.013 0.007 0.016 0.300 0.057 0.679 0.061 Ethereum
0.25 0.655 0.970 0.932 0.120 0.021 0.003 0.749
0.5 0.947 0.775 0.726 0.818 0.054 0.051 0.439
0.75 0.502 0.101 0.707 0.125 0.254 0.003 0.298
0.95 0.088 0.842 0.340 0.077 0.622 0.489 0.085
0.05 0.761 0.111 0.026 0.319 0.058 0.626 0.254 XRP
0.25 0.972 0.499 0.207 0.733 0.038 0.004 0.461
0.5 0.071 0.502 0.509 0.047 0.044 0.006 0.893
0.75 0.828 0.651 0.568 0.198 0.106 0.001 0.202
0.95 0.323 0.703 0.866 0.339 0.903 0.128 0.458
0.05 0.008 0.763 0.005 0.000 0.014 0.016 0.540 Tether
0.25 0.635 0.826 0.466 0.517 0.463 0.844 0.647
0.5 0.979 0.881 0.718 0.965 0.544 0.846 0.922
0.75 0.563 0.226 0.763 0.411 0.428 0.658 0.990
0.95 0.888 0.001 0.113 0.157 0.362 0.934 0.154
0.05 0.000 0.003 0.283 0.052 0.019 0.944 0.114 Bitcoin Cash
0.25 0.523 0.399 0.494 0.562 0.003 0.009 0.490
0.5 0.347 0.818 0.295 0.531 0.019 0.003 0.748
0.75 0.003 0.153 0.841 0.430 0.149 0.006 0.864
0.95 0.000 0.002 0.385 0.074 0.116 0.068 0.043
Next, we plotted the quantile coefficient variations for all variables for each cryptocur-
rency, as shown in Figure 3.
Figure 3 shows that Google trend searches have minimal loading among all factors,
followed by the global COVID-19 index changes. For the Bitcoin, Ethereum, and XRP
cryptocurrencies, Google trend searches were negatively related to the average returns
at the lower quartiles, whereas it became more positive at the higher quartile. Only
Bitcoin Cash shows a negative coefficient for Google trend searches at the highest quartile
(0.95) and it is statistically significant at the 10% level. Interestingly, in the case of Tether,
Google Trends search coefficients are statistically insignificant for all quantiles. The COVID-
19 coefficient is statistically significant (10%) at the lower tail for Ethereum and at the
higher tail in the case of Bitcoin. In the case of Bitcoin Cash, the COVID-19 coefficient is
statistically significant (5%) at both tails. Gold coefficients vary both in magnitude and
direction significantly across the quantiles for all five cryptocurrencies. Oil coefficients
are negative in all quartiles for Tether, with the negative value only in the highest (95th)
quartile for all other cryptocurrencies. S&P 500 coefficients are positive and statistically
significant in most of the cases. In the cases of Ethereum and XRP, the S&P 500 coefficients
are statistically insignificant at the two extreme ends of the tail. Conversely, in the case of
Bitcoin and Tether, the S&P 500 coefficients are statistically insignificant at the higher end
of the tail. In the case of Bitcoin Cash, S&P 500 coefficients are statistically significant at the
higher quartile, except for the extreme lower end of the tail (5th quartile). VIX coefficients
are statistically insignificant in all quartiles for two cryptocurrencies, namely XRP and
Tether. The VIX coefficient is statistically significant at the 5% level at the extreme lower
end (5th quartile) of Bitcoin and the highest (95th quartile) end of Bitcoin Cash, whereas, in
the case of Ethereum, the VIX coefficient is statistically significant at the 5% level at both
the extreme ends (5th and 95th quartiles).
0.5 0.347 0.818 0.295 0.531 0.019 0.003 0.748
0.75 0.003 0.153 0.841 0.430 0.149 0.006 0.864
0.95 0.000 0.002 0.385 0.074 0.116 0.068 0.043
Sustainability 2021, 13, 8578

Next, we plotted the quantile coefficient variations for all variables for 14each
of 17
cryptocurrency, as shown in Figure 3.
Sustainability 2021, 13, x FOR PEER REVIEW
Figure 3. Quantile coefficient variations.

Figure 3. Quantile coefficient variations.
The quantile regression appears to be3 ashows

Figure confirmation of thetrend
that Google key findings
searchesofhave
the previous
minimal loading
OLS estimates, with some improvements.
followed by the It isglobal
evident by looking
COVID-19 at the
index influence
changes. ForoftheS&P Bitcoin, Et
500 on the crypto market that the situation only
cryptocurrencies, slightly
Google trend differs across
searches were thenegatively
quartiles.related
For to the
example, at the lower quartile, this influence
the lower quartiles,iswhereas
highly strong on Bitcoin
it became and it remains
more positive at the higher qua
strong at the higher quartile. It is evident that Google trends and the
Cash shows a negative coefficient for Google trend searches COVID-19 crisisat the hig
(represented by the global COVID-19 index) have minimal loading effects on the crypto
and it is statistically significant at the 10% level. Interestingly, in the cas
market, with slight changesTrends
in thesearchmagnitude and direction
coefficients of coefficients
are statistically across
insignificant forthe
all quantil
quantile. Oil coefficients arecoefficient
negative in all quartilessignificant
is statistically for Tether,(10%)
whereas
at theoil coefficients
lower tail for Ethereum
are negative only in the highest
tail (95th)
in the quartile
case of for all other
Bitcoin. In cryptocurrencies.
the case of Bitcoin It is clearly
Cash, the COVID
confirmed, once again, that statistically
Tether’s behavior is very
significant (5%)specific
at bothintails.
the crypto market, as vary
Gold coefficients it both
shows some characteristics direction
of a safe haven withinacross
significantly a market. Additionally,
the quantiles our cryptocurrencies.
for all five findings
differ from those of [7]. We cannot
negative find inany
all safe havenfor
quartiles characteristics
Tether, with(only Tether isvalue
the negative near) only in
in comparison to their workquartile
and findings of diversification benefits and digital assets
for all other cryptocurrencies. S&P 500 coefficients are positiv
(social media) acting as safe significant
havens. in most of the cases. In the cases of Ethereum and XRP, the S
Additionally, the presentarestudy deployed
statistically robust regression
insignificant at the two with M-estimations.
extreme ends of theThe tail. Conver
M-estimation addresses dependent
Bitcoin and variable
Tether, outliers,
the S&Pwhere the value are
500 coefficients of the dependent
statistically insignifican
variable differs markedly from the tail.
of the regression model
In the case of norm
Bitcoin(large
Cash,residuals). The obtained
S&P 500 coefficients are statisti
the higher quartile, except for the extreme lower end of the tail (
coefficients are statistically insignificant in all quartiles for two cryptoc
XRP and Tether. The VIX coefficient is statistically significant at the 5% l
lower end (5th quartile) of Bitcoin and the highest (95th quartile) en
estimates of the robust regression are presented in Table 5. The obtained robust regression
estimates confirm the robustness of the OLS estimates except for the Tether. The effect on
the coefficient estimates of moving from least squares to robust M-estimation (see Tables 3
and 5) does not produce many distinct coefficient values (that are statistically significant at
the 5% level) than the OLS estimates, except for Tether.
Table 5. Robust regression estimates.
C COVID-19 GOLD GT OIL S&P 500 VIX R2

Bitcoin 0.059 −0.002 −0.055 −0.016 0.142 * 0.656 * 0.045 0.229
Ethereum 0.016 −0.001 0.160 −0.003 0.231 * 0.869 * 0.049 0.138
XRP −0.117 0.002 −0.175 0.033 0.148 * 0.730 * 0.027 0.225
Tether 0.000 0.000 0.007 0.000 0.001 0.018 * 0.002 0.031
Bitcoin Cash −0.091 −0.001 −0.521 0.024 0.216 * 1.097 * 0.060 0.194
5. Conclusions
Risk and returns are inseparable from each other in the financial markets [66–69]. In
terms of the impact of the first wave, gold and oil, as typical global commodities, could have
been diversifiers, with some characteristics of gold as a hedging instrument against risky
assets represented by S&P 500. The OLS, quantile, and robust regressions estimates show
that there was no significant direct influence of the COVID-19 crisis on the crypto market.
However, it seems there were spillovers from risky assets on the crypto market. Looking at
S&P 500 as the typical representative of risky assets on the global market and its influence
on the crypto market, there was a strong direct influence on all the cryptocurrencies except
Tether. In this way, it is largely clear that the crypto market cannot be either a safe haven
or hedge in this ongoing COVID-19 crisis. On the other hand, all analyses that have been
used (OLS, quantile regression, and robust regression estimates) have confirmed the very
specific behavior of Tether. The wavelet coherence analysis shows that only in crypto
pairs (which include Tether), there is a low level of coherence, with no contagion effect. It
appears that Tether is the only important stable coin in the crypto market and because of
this characteristic, it can present a safe haven within a crypto market. Its role in the crypto
market seems to be very similar to the role of gold as a typical global commodity on the
global market.
Even though the results of our study differ from those of [7], in that only Tether has
near safe haven characteristics (while the other cryptocurrencies cannot be recognized),
such studies have academic importance in analyzing assets in terms of major crises, and
providing methodological, theoretical, and applicative experience in investment finance
issues. Moreover, policymakers and investors cannot accept cryptocurrencies (especially
Bitcoin) as safe havens, but only as highly volatile and speculative assets (which is in line
with the work of [10,16,44]). Finally, the study does not come to an ultimate consensus on
whether the ongoing COVID-19 crisis is a “Black Swan” event [1–3,14] or not. Since it is
too early to comment on it, it is left for future studies to further examine the issue.
Author Contributions: Conceptualization, D.V., M.M. and Z.G.; formal analysis, D.V., M.M. and
Z.G.; investigation, D.V., M.M. and Z.G.; writing—original draft preparation, D.V., M.M. and Z.G.;
writing—review and editing, D.V., M.M., Z.G., E.M.G. and M.F.; funding acquisition, D.V., M.M.,
Z.G., E.M.G. and M.F. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by the RUDN University Strategic Academic Leadership Program.
Data Availability Statement: Publicly available datasets were analyzed in this study. This data can
be found as mentioned in Section 3.1 of the study.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. Grobys, K. When Bitcoin has the flu: On Bitcoin’s performance to hedge equity risk in the early wake of the COVID-19 outbreak.
Appl. Econ. Lett. 2021, 28, 860–865. [CrossRef]
2. Taleb, N. The Black Swan: The Impact of the Highly Improbable; Random House Inc.: New York, NY, USA, 2007.
3. Mnif, E.; Salhi, B.; Jarboui, A. Herding behaviour and Islamic market efficiency assessment: Case of Dow Jones and Sukuk market.
Int. J. Islam. Middle East. Financ. Manag. 2019, 13, 24–41. [CrossRef]
4. Vukovic, D.; Lapshina, K.A.; Maiti, M. European Monetary Union bond market dynamics: Pre & post crisis. Res. Int. Bus. Financ.
2019, 50, 369–380. [CrossRef]
5. Vukovic, D.; Vyklyuk, Y.; Matsiuk, N.; Maiti, M.; Chernova, N. Neural network forecasting in prediction Sharpe ratio: Evidence
from EU debt market. Phys. A Stat. Mech. Its Appl. 2020, 542, 123331. [CrossRef]
6. Mnif, E.; Jarboui, A.; Mouakhar, K. How the cryptocurrency market has performed during COVID 19? A multifractal analysis.
Financ. Res. Lett. 2020, 36, 101647. [CrossRef]
7. Corbet, S.; Hou, G.; Yang, H.; Lucey, B.M.; Les, O. Aye Corona! The Contagion Effects of Being Named Corona during the
COVID-19 Pandemic. 2020. Available online: https://ssrn.com/abstract=3561866 (accessed on 26 March 2020).
8. Jana, R.K.; Das, D. Did Bitcoin act as an antidote to the Chinese equity market and booster to Altcoins during the Novel
Coronavirus outbreak? SSRN 2020. [CrossRef]
9. Shahzad, S.J.H.; Bouri, E.; Roubaud, D.; Kristoufek, L.; Lucey, B. Is Bitcoin a better safe-haven investment than gold and
commodities? Int. Rev. Financ. Anal. 2019, 63, 322–330. [CrossRef]
10. Klein, T.; Thu, H.P.; Walther, T. Bitcoin is not the New Gold–A comparison of volatility, correlation, and portfolio performance.
Int. Rev. Financ. Anal. 2018, 59, 105–116. [CrossRef]
11. Das, D.; Le Roux, C.; Jana, R.; Dutta, A. Does Bitcoin hedge crude oil implied volatility and structural shocks? A comparison with
gold, commodity and the US Dollar. Financ. Res. Lett. 2020, 36, 101335. [CrossRef]
12. Corbet, S.; Lucey, B.; Urquhart, A.; Yarovaya, L. Cryptocurrencies as a financial asset: A systematic analysis. Int. Rev. Financ. Anal.
2019, 62, 182–199. [CrossRef]
13. Özer, M.; Kamenković, S.; Grubišić, Z. Frequency domain causality analysis of intra- and inter-regional return and volatility
spillovers of South-East European (SEE) stock markets. Econ. Res. Ekon. Istraživanja 2019, 33, 1–25. [CrossRef]
14. Yarovaya, L.; Matkovskyy, R.; Jalan, A. The COVID-19 Black Swan Crisis: Reaction and recovery of various financial markets.
SSRN 2020. [CrossRef]
15. Aalborg, H.A.; Molnar, P.; de Vries, J.E. What can explain the price, volatility and trading volume of Bitcoin? Financ. Res. Lett.
2020, 29, 255–265. [CrossRef]
16. Goodell, J.W. COVID-19 and finance: Agendas for future research. Financ. Res. Lett. 2020, 35, 101512. [CrossRef] [PubMed]
17. Goodell, J.W.; Goutte, S. Co-Movement of COVID-19 and Bitcoin: Evidence from Wavelet Coherence Analysis. 2020. Available
online: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3597144 (accessed on 1 June 2020).
18. Cheach, F. Speculative bubbles in Bitcoin markets? An empirical investigation into the fundamental value of Bitcoin. Econ. Lett.
2015, 130, 32–36. [CrossRef]
19. Hafner, C.M. Testing for bubbles in cryptocurrencies with time-varying volatility. J. Financ. Econ. 2018. [CrossRef]
20. Maiti, M.; Grubisic, Z.; Vukovic, D.B. Dissecting tether’s nonlinear dynamics during Covid-19. J. Open Innov. Technol. Mark.
Complex. 2020, 6, 161. [CrossRef]
21. Ardia, D.; Bluteau, K.; Rüede, M. Regime changes in Bitcoin GARCH volatility dynamics. Financ. Res. Lett. 2019, 29, 266–271.
[CrossRef]
22. Griffin, J.M.; Shams, A. Is bitcoin really un-tethered. SSRN 2019. [CrossRef]
23. Liu, Y.; Tsyvinski, A. Risks and returns of cryptocurrency. Rev. Financ. Stud. 2021, 34, 2689–2727. [CrossRef]
24. Brunnermeier, M.K.; Nagel, S. Hedge funds and the technology bubble. J. Financ. 2004, 59, 2013–2040. [CrossRef]
25. Shi, S.; Arora, V. An application of models of speculative behaviour to oil prices. Econ. Lett. 2012, 115, 469–472. [CrossRef]
26. Tsvetanov, D.; Coakley, J.; Kellard, N. Bubbling over! The behaviour of oil futures along the yield curve. J. Empir. Financ. 2016, 38,
516–533. [CrossRef]
27. Kruse, R.; Wegener, C. Time-varying persistence in real oil prices and its determinant. Energy Econ. 2020, 85, 104328. [CrossRef]
28. Liu, J.; Serletis, A. Volatility in the cryptocurrency market. Open Econ. Rev. 2019, 30, 779–811. [CrossRef]
29. Fakhfekh, M.; Jeribi, A. Volatility dynamics of crypto-currencies’ returns: Evidence from asymmetric and long memory GARCH
models. Res. Int. Bus. Financ. 2020, 51, 101075. [CrossRef]
30. Bouri, E.; Vo, X.V.; Saeed, T. Return equicorrelation in the cryptocurrency market: Analysis and determinants. Financ. Res. Lett.
2021, 38, 101497. [CrossRef]
31. Bouri, E.; Shahzad, S.J.H.; Roubaud, D. Co-explosivity in the cryptocurrency market. Financ. Res. Lett. 2019, 29, 178–183.
[CrossRef]
32. Baek, C.; Elbeck, M.A. Bitcoins as an investment or speculative vehicle? A first look. Appl. Econ. Lett. 2015, 22, 30–34. [CrossRef]
33. Dyhrberg, A.H. Bitcoin, gold and the dollar—A GARCH volatility analysis. Financ. Res. Lett. 2016, 16, 85–92. [CrossRef]
34. Bouri, E.; Das, M.; Gupta, R.; Roubaud, D. Spillovers between Bitcoin and other assets during bear and bull markets. Appl. Econ.
2018, 50, 5935–5949. [CrossRef]
35. Dyhrberg, A.H. Hedging capabilities of bitcoin. Is it a virtual gold. Financ. Res. Lett. 2016, 16, 139–144. [CrossRef]
36. Baur, D.G.; Dimpfl, T.; Kuck, K. Bitcoin, gold and the US dollar—A replication and extension. Financ. Res. Lett. 2018, 25, 103–110.
[CrossRef]
37. Jeevan, R.; Shubhashree, P.K.A. Arbitrary behavior of bitcoin towards gold and crudeoil. Int. J. Res. Anal. Rev. 2019, 6, 481–486.
38. Baur, D.G.; Glover, K.J. The destruction of a safe haven asset? Appl. Financ. Lett. 2012, 1, 8. [CrossRef]
39. Shrydeh, N.; Shahateet, M.; Mohammad, S.; Sumadi, M. The hedging effectiveness of gold against US stocks in a post-financial
crisis era. Cogent Econ. Financ. 2019, 7. [CrossRef]
40. Chen, K.; Wang, M. Is gold a hedge and safe haven for stock market? Appl. Econ. Lett. 2018, 26, 1080–1086. [CrossRef]
41. Dar, A.B.; Mitra, D. Is gold a weak or strong hedge and safe haven against stocks? Robust evidences from three major gold-
consuming countries. Appl. Econ. 2017, 49, 5491–5503. [CrossRef]
42. Beckmann, J.; Berger, T.; Czudaj, R. Gold price dynamics and the role of uncertainty. Quant. Financ. 2018, 19, 663–681. [CrossRef]
43. Maiti, M.; Vyklyuk, Y.; Vuković, D. Cryptocurrencies chaotic co-movement forecasting with neural networks. Internet Technol.
Lett. 2020, 3, e157. [CrossRef]
44. Qiuhua, X.; Yixuan, Z.; Ziyang, Z. Tail-risk spillovers in cryptocurrency markets. Financ. Res. Lett. 2021, 38, 101453.
45. Ji, Q.; Bouri, E.; Lau, C.K.M.; Roubaud, D. Dynamic connectedness and integration in cryptocurrency markets. Int. Rev. Financ.
Anal. 2019, 63, 257–272. [CrossRef]
46. Zi˛eba, D.; Kokoszczynski, R.; Śledziewska, K. Shock transmission in the cryptocurrency market. Is Bitcoin the most influential?
Int. Rev. Financ. Anal. 2019, 64, 102–125. [CrossRef]
47. Härdle, W.K.; Wang, W.; Yu, L. TENET: Tail-Event driven NETwork risk. J. Econ. 2016, 192, 499–513. [CrossRef]
48. Fan, Y.; Härdle, W.K.; Wang, W.; Zhu, L. Single-index-based CoVaR with very high-dimensional covariates. J. Bus. Econ. Stat.
2017, 36, 212–226. [CrossRef]
49. Ike, G.N.; Usman, O.; Sarkodie, S.A. Testing the role of oil production in the environmental Kuznets curve of oil producing
countries: New insights from Method of Moments Quantile Regression. Sci. Total Environ. 2020, 711, 135208. [CrossRef]
50. Koenker, R.W.; Bassett, G., Jr. Regression quantiles. Econometrica 1978, 46, 33–50. [CrossRef]
51. Koenker, R.W.; Hallock, G. Quantile regression. J. Econ. Perspect. 2001, 15, 143–156. [CrossRef]
52. Allen, D.E.; Powell, S.R. Asset pricing, the Fama—French Factor Model and the implications of quantile-regression analysis. In
Financial Econometrics Modeling: Market Microstructure, Factor Models and Financial Risk Measures; Gregoriou, G.N., Pascalau, G.N.,
Eds.; Palgrave Macmillan: London, UK, 2011.
53. Barnes, M.L.; Hughes, A.W. A Quantile regression analysis of the cross section of stock market returns. SSRN 2004. [CrossRef]
54. Onyedikachi, J. Quantile regression analysis as a robust alternative to ordinary least squares. Sci. Afr. 2009, 8, 61–65.
55. Karlsson, A. Estimation and inference for quantile regression of longitudinal data, with applications in biostatistic. Diss. Acta
Univ. Ups. 2006, 18, 36.
56. Buchinsky, M. Recent advances in quantile regression models: A practical guideline for empirical research. J. Hum. Resour. 1998,
33, 88. [CrossRef]
57. Jareño, F.; de la O González, M.; Tolentino, M.; Sierra, K. Bitcoin and gold price returns: A quantile regression and NARDL
analysis. Resour. Policy 2020, 67, 101666. [CrossRef]
58. Nardo, M.; Saisana, M.; Saltelli, A.; Tarantola, S. Tools for Composite Indicators Building; Working paper No. EUR 21682 EN.;
European Commission: Ispra, Italy, 2005.
59. Nardo, M.; Saisana, M.; Saltelli, A.; Tarantola, S.; Hoffman, A.; Giovannini, E. Handbook on Constructing Composite Indicators:
Methodology and User Guide; OECD: Ispra, Italy, 2008.
60. Maiti, M.; Jadhav, P. Impact of pollution level, death rate and illness on economic growth: Evidences from global economy. SN
Bus. Econ. 2021. [CrossRef]
61. Vukovic, D.B.; Lapshina, K.A.; Maiti, M. Wavelet coherence analysis of returns, volatility and interdependence of the US and the
EU money markets: Pre & post crisis. N. Am. J. Econ. Financ. 2021, 58, 101457. [CrossRef]
62. Maiti, M.; Vukovic, D.; Krakovich, V.; Pandey, M.K. How integrated are cryptocurrencies. Int. J. Big Data Manag. 2020, 1, 64–80.
[CrossRef]
63. Maiti, M. OLS versus quantile regression in extreme distributions. Contaduría Adm. 2019, 64, 12. [CrossRef]
64. Maiti, M. Quantile regression, asset pricing and investment decision. IIMB Manag. Rev. 2021, 33, 28–37. [CrossRef]
65. Maiti, M.; Maiti, J. A study on India’s first offshore LIDAR-based wind profiling at Gulf of Khambhat. J. Public Aff. 2020, 20, e2044.
[CrossRef]
66. Maiti, M. A critical review on evolution of risk factors and factor models. J. Econ. Surv. 2020, 34, 175–184. [CrossRef]
67. Maiti, M. A Six Factor Asset Pricing Model. Ph.D. Thesis, Pondicherry University, Puducherry, India, 2018. Available online:
http://dspace.pondiuni.edu.in/jspui/bitstream/123456789/3180/1/T6617.pdf (accessed on 1 May 2021).
68. Maiti, M. Is ESG the succeeding risk factor? J. Sustain. Financ. Invest. 2021, 11, 199–213. [CrossRef]
69. Maiti, M.; Balakrishnan, A. Can leverage effect coexist with value effect? IIMB Manag. Rev. 2020, 32, 7–23. [CrossRef]

Sustainability 13 08578 v5

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sustainability 13 08578 v5

Uploaded by

Copyright:

Available Formats

sustainability

Received: 25 June 2021

Sustainability 2021, 13, 8578. https://doi.org/10.3390/su13158578 https://www.mdpi.com/journal/sustainability

3. Data and Methodology

COVID VIX GOLD OIL S&P 500

Apr 01, 2020

Bitcoin Etherum XRP BitcoinCash Tether

Sustainability 2021, R it = a + b COVID-19 + s GOLDt + gterm;

Rp, d,b,and s, g,f are p, d,the and sensitivity

4. Results and Discussion

Table 1. Descriptive statistics.

COVID-19 GOLD GT * OIL S&P 500 VIX

Table 2. Correlation analysis.

COVID-19 GOLD GT * OIL S&P 500 VIX

4.2. Wavelet Coherence

Figure 2. Wavelet coherence diagram among the cryptocurrency pairs.

4.3. OLS Estimates

Table 3. OLS regression estimates.

C COVID-19 GOLD GT OIL S&P 500 VIX R2

4.4. Quantile Regression Estimates

Table 4. Quantile regression estimates.

0.95 0.316 0.065 0.595 0.076 0.293 0.375 0.966

Sustainability 2021, 13, 8578

cryptocurrency, as shown in Figure 3.

Sustainability 2021, 13, x FOR PEER REVIEW

Figure 3. Quantile coefficient variations.

The quantile regression appears to be3 ashows

Table 5. Robust regression estimates.

C COVID-19 GOLD GT OIL S&P 500 VIX R2

You might also like